Spark Streaming Large-scale near-real-time stream processing

1. Spark StreamingLarge-scale near-real-time stream processing • Tathagata Das (TD) • along with • MateiZaharia, Haoyuan Li, Timothy Hunter, Scott Shenker, Ion Stoica, and many others UC BERKELEY

2. What is Spark Streaming? • Extends Spark for doing large scale stream processing • Scales to 100s of nodes and achieves second scale latencies • Efficient and fault-tolerant stateful stream processing • Simple batch-like API for implementing complex algorithms

3. Motivation • Many important applications must process large streams of live data and provide results in near-real-time • Social network trends • Website statistics • Ad impressions … • Distributed stream processing framework is required to • Scale to large clusters (100s of machines) • Achieve low latency (few seconds)

4. Integration with Batch Processing • Many environments require processing same data in live streaming as well as batch post processing • Existing framework cannot do both • Either do stream processing of 100s of MB/s with low latency • Or do batch processing of TBs / PBs of data with high latency • Extremely painful to maintain two different stacks • Different programming models • Double the implementation effort • Double the number of bugs

5. StatefulStream Processing • Traditional streaming systems have a record-at-a-time processing model • Each node has mutable state • For each record, update state and send new records • State is lost if node dies! • Making stateful stream processing be fault-tolerant is challenging mutable state input records node 1 node 3 input records node 2

6. Existing Streaming Systems • Storm • Replays record if not processed by a node • Processes each record at least once • May update mutable state twice! • Mutable state can be lost due to failure! • Trident – Use transactions to update state • Processes each record exactly once • Per state transaction to external database is slow

7. Spark Streaming

8. Discretized Stream Processing Run a streaming computation as a series of very small, deterministic batch jobs live data stream Spark Streaming • Chop up the live stream into batches of X seconds • Spark treats each batch of data as RDDs and processes them using RDD operations • Finally, the processed results of the RDD operations are returned in batches batches of X seconds Spark processed results

9. Discretized Stream Processing Run a streaming computation as a series of very small, deterministic batch jobs live data stream Spark Streaming • Batch sizes as low as ½ second, latency of about 1 second • Potential for combining batch processing and streaming processing in the same system batches of X seconds Spark processed results

10. Example – Get hashtags from Twitter valtweets= ssc.twitterStream() DStream: a sequence of RDDs representing a stream of data Twitter Streaming API tweets DStream batch @ t+1 batch @ t batch @ t+2 stored in memory as an RDD (immutable, distributed)

11. Example – Get hashtags from Twitter val tweets = ssc.twitterStream() valhashTags= tweets.flatMap(status => getTags(status)) new DStream transformation: modify data in one DStreamto create another DStream tweets DStream batch @ t+1 batch @ t batch @ t+2 hashTagsDstream [#cat, #dog, … ] new RDDs created for every batch flatMap flatMap flatMap …

12. Example – Get hashtags from Twitter val tweets = ssc.twitterStream() valhashTags = tweets.flatMap (status => getTags(status)) hashTags.saveAsHadoopFiles("hdfs://...") output operation: to push data to external storage batch @ t+1 batch @ t batch @ t+2 tweets DStream flatMap flatMap flatMap hashTags DStream every batch saved to HDFS save save save

13. Example – Get hashtags from Twitter val tweets = ssc.twitterStream() valhashTags = tweets.flatMap (status => getTags(status)) hashTags.foreach(hashTagRDD => { ... }) foreach: do whatever you want with the processed data batch @ t+1 batch @ t batch @ t+2 tweets DStream flatMap flatMap flatMap hashTags DStream foreach foreach foreach Write to database, update analytics UI, do whatever you want

14. Java Example Scala valtweets= ssc.twitterStream() valhashTags= tweets.flatMap(status => getTags(status)) hashTags.saveAsHadoopFiles("hdfs://...") Java JavaDStream<Status> tweets = ssc.twitterStream() JavaDstream<String> hashTags= tweets.flatMap(new Function<...> { }) hashTags.saveAsHadoopFiles("hdfs://...") Function object

15. Window-based Transformations val tweets = ssc.twitterStream() valhashTags = tweets.flatMap (status => getTags(status)) valtagCounts = hashTags.window(Minutes(1), Seconds(5)).countByValue() sliding window operation window length sliding interval window length sliding interval DStream of data

16. Arbitrary Stateful Computations Specify function to generate new state based on previous state and new data • Example: Maintain per-user mood as state, and update it with their tweets updateMood(newTweets, lastMood) => newMood moods = tweets.updateStateByKey(updateMood _)

17. Arbitrary Combinations of Batch and Streaming Computations Inter-mix RDD and DStream operations! • Example: Join incoming tweets with a spam HDFS file to filter out bad tweets tweets.transform(tweetsRDD => {tweetsRDD.join(spamHDFSFile).filter(...) })

18. DStream Input Sources • Out of the box we provide • Kafka • HDFS • Flume • Akka Actors • Raw TCP sockets • Very easy to write a receiver for your own data source

19. Fault-tolerance: Worker • RDDs remember the operations that created them • Batches of input data are replicated in memory for fault-tolerance • Data lost due to worker failure, can be recomputed from replicated input data tweets RDD input data replicated in memory flatMap hashTags RDD lost partitions recomputed on other workers • All transformed data is fault-tolerant, and exactly-once transformations

20. Fault-tolerance: Master • Master saves the state of the DStreams to a checkpoint file • Checkpoint file saved to HDFS periodically • If master fails, it can be restarted using the checkpoint file • More information in the Spark Streaming guide • Link later in the presentation • Automated master fault recovery coming soon

21. Performance Can process 6 GB/sec (60M records/sec) of data on 100 nodes at sub-second latency • Tested with 100 text streams on 100 EC2 instances with 4 cores each

22. Comparison with Storm and S4 Higher throughput than Storm • Spark Streaming: 670k records/second/node • Storm: 115k records/second/node • Apache S4: 7.5k records/second/node

23. Fast Fault Recovery Recovers from faults/stragglers within 1 sec

24. Real Applications: Mobile Millennium Project Traffic transit time estimation using online machine learning on GPS observations • Markov chain Monte Carlo simulations on GPS observations • Very CPU intensive, requires dozens of machines for useful computation • Scales linearly with cluster size

25. Real Applications: Conviva Real-time monitoring and optimization of video metadata • Aggregation of performance data from millions of active video sessions across thousands of metrics • Multiple stages of aggregation • Successfully ported to run on Spark Streaming • Scales linearly with cluster size

26. Unifying Batch and Stream Processing Models Spark program on Twitter log file using RDDs valtweets= sc.hadoopFile("hdfs://...") valhashTags= tweets.flatMap(status => getTags(status)) hashTags.saveAsHadoopFile("hdfs://...") Spark Streaming program on Twitter stream using DStreams valtweets= ssc.twitterStream() valhashTags= tweets.flatMap(status => getTags(status)) hashTags.saveAsHadoopFiles("hdfs://...")

27. Vision - one stack to rule them all $ ./spark-shell scala> val file = sc.hadoopFile(“smallLogs”) ... scala> val filtered = file.filter(_.contains(“ERROR”)) ... scala> val mapped = filtered.map(...) ... • Explore data interactively using Spark Shell to identify problems • Use same code in Spark stand-alone programs to identify problems in production logs • Use similar code in Spark Streaming to identify problems in live log streams object ProcessProductionData { def main(args: Array[String]) { valsc = new SparkContext(...) val file = sc.hadoopFile(“productionLogs”) val filtered = file.filter(_.contains(“ERROR”)) val mapped = filtered.map(...) ... } } object ProcessLiveStream { def main(args: Array[String]) { valsc = new StreamingContext(...) val stream = sc.kafkaStream(...) val filtered = file.filter(_.contains(“ERROR”)) val mapped = filtered.map(...) ... } }

28. Vision - one stack to rule them all Spark + Shark + Spark Streaming

29. Today’s Tutorial • Process Twitter data stream to find most popular hashtags • Requires a Twitter account • Need to setup Twitter OAuth keys • All the instructions in the tutorial • Your account is safe! • No need to enter your password anywhere, only enter the keys in configuration file • Destroy the keys after the tutorial is done

30. Conclusion • Integrated with Spark as an extension • Takes 5 minutes to spin up a Spark cluster to try it out • Streaming programming guide – http://spark.incubator.apache.org/docs/latest/streaming-programming-guide.html • Paper – tinyurl.com/dstreams Thank you!